OpenAI Releases Findings on 377 Math Problems, Further Roiling Field

OpenAI has released findings involving 377 mathematics problems, according to reporting by The New York Times. The release has added to debate in the artificial-intelligence field over how advanced systems should be evaluated and what their performance on difficult tasks demonstrates.

Math problems are commonly used in AI research as one way to test reasoning, because they can have defined answers and may require multiple steps to solve. But results can be interpreted differently depending on how problems are selected, presented to a model and scored.

OpenAI is a prominent developer of generative AI systems, and its research announcements are closely watched by companies, academics and policymakers. The findings on the 377 problems have further roiled a field already divided over AI capabilities, measurement and the implications of increasingly capable models.